Patient-month modeling for longitudinal blood culture contamination surveillance during a major healthcare-system disruption: a single-center interrupted time-series study.

INTRODUCTION

Blood culture contamination (BCC) is an important diagnostic quality indicator, but crude temporal comparisons may be misleading when repeated sampling within patients is not considered. We evaluated longitudinal changes in BCC rates before, during, and after the COVID-19 pandemic using patient-month Poisson generalized estimating equations (GEE).

METHODS

This single-center retrospective study included blood culture sets obtained at a tertiary hospital in Japan from April 2017 to June 2025. Eligible episodes had at least two sets drawn within 24 h. Interrupted time-series analysis estimated incidence rate ratios (IRRs) for level and slope changes at pandemic and post-reclassification transitions, with the number of sets as an offset and adjustment for seasonality and within-patient correlation.

RESULTS

Among 85,955 blood culture sets, 84,213 sets from 41,732 episodes in 23,705 patients were analyzed. Crude BCC rates were 1.05%, 1.55%, and 1.49% in the prepandemic, pandemic, and post-reclassification periods. No significant overall level or slope changes were observed: prepandemic-to-pandemic level-change IRR, 1.23 (95% CI, 0.93-1.62); slope-change IRR, 1.00 (95% CI, 0.99-1.01); pandemic-to-post-reclassification level-change IRR, 0.88 (95% CI, 0.68-1.14); and slope-change IRR, 0.99 (95% CI, 0.98-1.01). Exploratory subgroup analyses suggested setting-specific post-reclassification changes in the intensive care unit. Sensitivity analyses were generally consistent, although the Omicron-era breakpoint analysis suggested sensitivity to breakpoint specification.

CONCLUSION

Patient-month Poisson GEE-based analysis showed that crude increases in BCC rates were not accompanied by statistically significant overall level or slope changes after accounting for repeated patient-level sampling and seasonality. Crude BCC trends should therefore be interpreted cautiously in longitudinal microbiology surveillance.